A graph-tool wrapper for handling large multi-layered networks
There are a number of different python packages for dealing with networks, each with their own pros and cons. NetworkX is probably the widest used in the python ecosystem, with a gentle learning curve and a vast array of functionality. But being a purely python implementation, it scales very poorly to large networks. For example in this blogpost, although somewhat aged now, benchmarking showed NetworkX performed nearly 100x slower than graph-tool on pagerank. While more efficient packages implemented in C++ with python wrappers are available, such as graph-tool, they may offer more targeted functionality or have a steeper learning curve to operate.
OnionNet is a wrapper for graph-tool to make it easier to build, manipulate, analyse and visualise very large multi-layered networks. This was spurred from my own needs faced in the creation and analysis of LipiNet, but the package is abstract enough to be potentially useful for anyone working large multilayered networks of any kind.
OnionNet first requires graph-tool to be installed. Because graph-tool is built around C++ for efficiency, unfortunately there is no straightforward pip installation. Nonetheless, there are a number of ways to install graph-tool besides pip, see here for more details. The easiest way for most users is probably to create a new env via conda:
conda create --name gt -c conda-forge graph-tool ipython jupyter
conda activate gt
Then you can install OnionNet with:
git clone https://github.com/saezlab/onionnet.git
cd onionnet
pip install -e .
In the near future we intend to include OnionNet on PyPI.
If you have pandas dataframes with nodes and edges, then getting started is quick and simple. See the getting_started.ipynb for a full walkthrough from scratch.
OnionNet is in active development and subject to change. Some functions may be modified or deprecated in future releases. If you find OnionNet helpful or have ideas for improvement, we'd love to hear more!